145 add support for constructing from data without copying (ktensor)

.gitignore

@@ -12,3 +12,4 @@ htmlcov
 __pycache__
 .vs*
 TestResults
+.DS_Store

pyttb/cp_als.py

@@ -68,15 +68,15 @@ def cp_als(
     >>> weights = np.array([1., 2.])
     >>> fm0 = np.array([[1., 2.], [3., 4.]])
     >>> fm1 = np.array([[5., 6.], [7., 8.]])
-    >>> K = ttb.ktensor.from_data(weights, [fm0, fm1])
+    >>> K = ttb.ktensor([fm0, fm1], weights)
     >>> np.random.seed(1)
     >>> M, Minit, output = ttb.cp_als(K.full(), 2) # doctest: +ELLIPSIS
     CP_ALS:
      Iter 0: f = ... f-delta = ...
      Iter 1: f = ... f-delta = ...
      Final f = ...
     >>> print(M) # doctest: +ELLIPSIS
-    ktensor of shape 2 x 2
+    ktensor of shape (2, 2)
     weights=[108.4715... 8.6114...]
     factor_matrices[0] =
     [[0.4187... 0.3989...]
@@ -85,7 +85,7 @@ def cp_als(
     [[0.6188... 0.2581...]
      [0.7854... 0.9661...]]
     >>> print(Minit) # doctest: +ELLIPSIS
-    ktensor of shape 2 x 2
+    ktensor of shape (2, 2)
     weights=[1. 1.]
     factor_matrices[0] =
     [[4.1702...e-01 7.2032...e-01]
@@ -147,12 +147,12 @@ def cp_als(
             factor_matrices.append(
                 np.random.uniform(0, 1, (input_tensor.shape[n], rank))
             )
-        init = ttb.ktensor.from_factor_matrices(factor_matrices)
+        init = ttb.ktensor(factor_matrices)
     elif isinstance(init, str) and init.lower() == "nvecs":
         factor_matrices = []
         for n in range(N):
             factor_matrices.append(input_tensor.nvecs(n, rank))
-        init = ttb.ktensor.from_factor_matrices(factor_matrices)
+        init = ttb.ktensor(factor_matrices)
     else:
         assert False, "The selected initialization method is not supported"
 
@@ -208,7 +208,7 @@ def cp_als(
             U[n] = Unew
             UtU[:, :, n] = U[n].T @ U[n]
 
-        M = ttb.ktensor.from_data(weights, U)
+        M = ttb.ktensor(U, weights)
 
         # This is equivalent to innerprod(X,P).
         iprod = np.sum(

pyttb/cp_apr.py

@@ -116,7 +116,7 @@ def cp_apr(
             factor_matrices.append(
                 np.random.uniform(0, 1, (input_tensor.shape[n], rank))
             )
-        init = ttb.ktensor.from_factor_matrices(factor_matrices)
+        init = ttb.ktensor(factor_matrices)
 
     # Call solver based on the couce of algorithm parameter, passing all the other input parameters
     if algorithm.lower() == "mu":
@@ -243,8 +243,7 @@ def tt_cp_apr_mu(
     nTimes = np.zeros((maxiters,))
 
     # Set up for iteration - initializing M and Phi.
-    # TODO replace with copy
-    M = ttb.ktensor.from_tensor_type(init)
+    M = init.copy()
     M.normalize(normtype=1)
     Phi = []  # np.zeros((N,))#cell(N,1)
     for n in range(N):
@@ -453,8 +452,7 @@ def tt_cp_apr_pdnr(
             init[n][tmpIdx, 0] = 1e-8
 
     # Start with the initial guess, normalized using the vector L1 norm
-    # TODO replace with copy
-    M = ttb.ktensor.from_tensor_type(init)
+    M = init.copy()
     M.normalize(normtype=1)
 
     # Sparse tensor flag affects how Pi and Phi are computed.
@@ -824,8 +822,7 @@ def tt_cp_apr_pqnr(
             init[n][tmpIdx, 0] = 1e-8
 
     # Start with the initial guess, normalized using the vector L1 norm
-    # TODO replace with copy
-    M = ttb.ktensor.from_tensor_type(init)
+    M = init.copy()
     M.normalize(normtype=1)
 
     # Sparse tensor flag affects how Pi and Phi are computed.

pyttb/import_data.py

@@ -31,14 +31,14 @@ def import_data(filename, index_base=1):
         shape = import_shape(fp)
         data = import_array(fp, np.prod(shape))
         fp.close()
-        return ttb.tensor().from_data(data, shape)
+        return ttb.tensor.from_data(data, shape)
 
     elif data_type == "sptensor":
         shape = import_shape(fp)
         nz = import_nnz(fp)
         subs, vals = import_sparse_array(fp, len(shape), nz, index_base)
         fp.close()
-        return ttb.sptensor().from_data(subs, vals, shape)
+        return ttb.sptensor.from_data(subs, vals, shape)
 
     elif data_type == "matrix":
         shape = import_shape(fp)
@@ -59,7 +59,7 @@ def import_data(filename, index_base=1):
             fac = np.reshape(fac, np.array(fac_shape))
             factor_matrices.append(fac)
         fp.close()
-        return ttb.ktensor().from_data(weights, factor_matrices)
+        return ttb.ktensor(factor_matrices, weights, copy=False)
 
 
 def import_type(fp):